By Timothy Prickett Morgan
The classic marketing scheme to use on any new and costly technology is to give away the expensive hardware and make up the sales volume – as well as revenues and profits – selling supplies or services. The phone company, knowing full well that early telephone users would not pay a high price for a telephone, gave away the box and sold per minute connection charges. In the early days of punch card computing, IBM not only had strict control of its punch card machines, but tweaked its machines so only IBM punch cards could be used in its devices.
The computer business has thus far been more or less immune from this kind of marketing approach because servers and PCs were relatively expensive pieces of capital equipment that needed lots of programming to make them useful and even more tender loving care to keep them running. But a number of different forces have come together in 1999 to make the idea of a free PC not only possible, but imminently practical for a large portion of the worldwide installed base of PC users. While much has been made of the nascent Free PC movement in the past few months as PC vendors such as Gateway and Dell have offered customers almost free PCs in exchange for their commitment to use their new internet services, or as companies such as Microworkz have committed to offering customers even less costly machines – under $200 – if they agree to sign up with them to get repackaged AT&T Worldnet Internet services. Earlier in the year, FreePC, perhaps the most innovative of companies applying the free approach to PCs, started giving away free PCs and internet access in exchange for personal information and submitting FreePC users to a barrage of advertising. This makes FreePC a kind of captive ISP, if you think about it.
While companies distributing free computers have been focusing on the PC market because of the high volumes of potential customers and the relatively low cost of the devices, it’s high time that some of the white box server vendors, or perhaps struggling Unix or PC server vendors should consider extending the focus to servers. While this may sound like an unworkable idea, especially given the fact that the big computer companies that sell both desktop and server computers get a big portion of their profits from servers, the Free Server idea bears some merit.
Regardless of what Intel likes to think, servers are definitely fast becoming commodities. (The irony is that this is happening just as Intel and its partners are finally getting ready to bring mainframe-class iron to market with the Profusion eight-way servers, due by the end of summer.) But just because servers are commodities, or soon will be, doesn’t mean companies can’t make money selling them, or in this case, by giving them away and selling some kind of service with them.
The application service providers or information utilities that most of the big server vendors envision as being their salvation are just a twist on this idea. But bringing up the idea may in the end be the undoing of IBM, Hewlett-Packard and Sun Microsystems, all of which have hardware and software architectures to protect and to proliferate. All customers will ultimately care about is that when they plug into their server, it speaks database and internet fluently. IBM’s recent Magic Box advertisements for its server line dance around this idea, but they miss the point that IBM has got to convince customers that it has very specific differentiators to survive against commodity server vendors. The ads, of course, don’t do this because IBM’s marketeers are trying to be cute and they seem to be rightfully stumped in coming up with a long list of feeds and speeds that make an RS/6000 or an AS/400 look better or a lot different from just about any other Unix server on the market. They are surely different when it comes to running old legacy code, but when you’re talking about using web application servers and Enterprise JavaBeans, the differences come down mainly to striking price and
buyer preference.
The server hardware’s share of a configured server set-up has been slipping steadily since Unix started being a commercial alternative in the early 1980s, and as memory and disk prices come down at the same rates as processor prices have in the last decade, server hardware will represent a very small piece of business compared to systems and applications software. In 1993, when IBM first shipped its DB2 data base on its RS/6000 line of servers, an RS/6000 model 970 server using a 50MHz Power1 processor and configured with 1Gb of memory and 50Gb of disk cost just over $500,000 including AIX V3 and DB2/6000 for about 300 online users, the practical maximum number of online transaction processing users on the box (that number is an estimate for TPC-C users). If you do the math, hardware accounted for two thirds of that set-up, and software the other third. The total cost came to about $1,500 per user for this machine.
In 1999, a four-way RS/6000 H70 server using 340MHz Northstar processors, 8Gb of memory, 145Gb of disk and AIX V4 and DB2 Universal Data Base (enterprise edition) costs about $370,000. The hardware accounts for only about half of the cost in this set-up, and mainly because even at $12 per Mb, 8Gb of memory comes to a fair chunk of change. If IBM has been selling DB2 on the RS/6000s at the same price when it announced the original models in early 1990, the fastest machine, the 930 (with a 25MHz Power1 chip) configured with 256Mb of memory and 12Gb of disk would have cost about $550,000 with half as many DB2 users, and hardware would have accounted for 80% of that total price tag. While this may not seem to be a dramatic shift in the allocation of hardware versus software dollars, the trend is pretty clear.
The other shocking thing about this comparison is that the four- way H70 can support approximately 14,000 TPC-C online users, about fifty times as many users as that old 970 machine and a hundred times as many as the older 930. This downward trend will not be so impressive on technical and engineering workloads because as hardware has advanced, CAD/CAM software has moved from 2D to 3D to realtime and has burned up all the MIPS IBM can cram in an RS/6000. For example, about 90% of the $475,000 a customer would have spend on a model 930 server in 1991 with five CATIA users went for hardware. (CATIA is the CAD/CAM software that France’s Dassault Systemes and IBM co-developed and which is one of the major drivers of the RS/6000 workstation and server business.) While the H70 is tuned for commercial workloads, it can nonetheless support CATIA, and about 80% of the price of buying an H70 with enough disk and memory to support two dozen CATIA users goes towards hardware.
Obviously, the percentage of money spent on hardware and software changes depending on the server, its configuration and the system software included, but one thing doesn’t change: as you factor in applications and middleware, hardware starts diminishing rather fast. Software modules in popular ERP packages cost hundreds of thousands of dollars, and full suites cost millions after code customization. Server costs get dwarfed in such comparisons, and that is something that none of the big server vendors even want customers to think about. The question is how a server vendor can use this to its advantage.
Clearly, any server vendor that wanted the Free Server movement to actually take off would have to gobble up ERP, SCM and CRM application suites so they could put them exclusively on their machines. Alternatively, an ERP vendor could simply start a hardware division or merge with an upstart server company like SGI or Gateway and give away free servers to customers that use its applications. Or a middleman could pre-configure two or three white box Unix or NT servers, slap on turnkey applications and make it up in volume. Any server vendor with nothing to lose has a great chance of taking away lots of business from IBM, Compaq, HP, Sun, Hitachi, Fujitsu and SGI – simply because none of these firms can afford to give away their servers to compete. Intel can’t do it – although it must be sorely tempted, as it mulls building servers for ISPs to steal some money directly from Sun’s back pocket – because it would alienate many of the very companies who buy the bulk of the Pentium III processors that end up in servers today and the Merced and McKinley processors that will be used in the coming years. Sooner or later, though, someone will probably figure out that you can make more money giving a server away (and thereby greasing the skids to slide in software and services) than trying to sell it directly.
